Solar wind turbulent spectrum at plasma kinetic scales

Draft version December 4, 2012

Preprint typeset using L£-T^X style cmulateapj v. 5/2/11

(N

o

(N
o

Q

SOLAR WIND TURBULENT SPECTRUM AT PLASMA KINETIC SCALES

O. Alexandrova 1 , C. Lacombe 1 , A. Mangeney 1 , R. Grappin 23 and M. Maksimovic 1

("Dated: November 15, 2012)
Draft version December 4, 2012

ABSTRACT

The description of the turbulent spectrum of magnetic fluctuations in the solar wind in the ki-
netic range of scales is not yet completely established. Here, we perform a statistical study of 100
spectra measured by the STAFF instrument on the Cluster mission, which allows to resolve tur-
bulent fluctuations from ion scales down to a fraction of electron scales, i.e. from ~ 10 2 km to
~ 300 m. We show that for k±p e e [0.03, 3] (that corresponds approximately to the frequency in
the spacecraft frame / £ [3,300] Hz), all the observed spectra can be described by a general law

E(k±) oc fc^ 8 ^ 3 exp (— k±p e ), where k± is the wave- vector component normal to the background mag-
netic field and p e the electron Larmor radius. This exponential tail found in the solar wind seems
compatible with the Landau damping of magnetic fluctuations onto electrons.
Subject headings: solar wind, plasma turbulence, kinetic scales

a:

a.

>

(N

o

X

1. INTRODUCTION

In neutral, homogeneous and isotropic fluids, the tur-
bulent fluctuations are unpredi ctable, but th eir statistics
are predictable and universal (|Frisch|[l995D ; the turbu-
lent spectra follow the power-law ~ fc~ 5 / 3 for any local
conditions (k being the wave numbe r). This empirica l
result was explained by Kolmogorov ()Kolmogorov||194ll)
assuming self similarity of the turbulent fluctuations be-
tween the energy injection scale and the dissipation one
U

In the magnetized solar wind, collisions are very rare
(the mean free path is of the order of 1 AU); the dissipa-
tion process at work and the dissipation length are not
known precisely. Moreover, in a magnetized plasma, it
is difficult to imagine self-similarity over all scales where
turbulent fluctuations are observed, since there exist sev-
eral spatial and temporal characteristic scales, such as
the ion Larmor radius pi — y / 2kTi±/mi/(2Trf C i), the
ion inertia length Xi = c/ui p i, the corresponding elec-
tron scales p e , A e , and the ion and electron cyclotron fre-
quencies fd, fee. At these scales, the dominant physical
processes change, which affects the scaling of the energy
transfer time and furthermore the energy transfer rate,
leading to spectral shape changes.

The first clear spectral change appears at ion scales. At
1 AU, the ion scales are nearly equal, A,; ~ pi ~ V/2Ttf C i,
so it is difficult to determine which of these scales is
responsible for the ion break. Independent measure-
ments at differe nt distances from the S un, between 0.3
AU to 0.9 AU (iBourouaine et al.ll2012f). and a statisti-
cal study at 1 AU (jLeamon et al.l 12000) indicate that
the spectral break is related to the ion inertia length
Xi. Nearly incompressible magnetic fluctuations cascad-
ing from the inertial range may undergo kinetic effects
in the vicinity of the ion scales. At these scales, ion

CNRS, UPMC Universite
5 place J. Janssen, 92190

1 LESIA-Obscrvatoire de Paris,
Paris 06, Universite Paris-Diderot
Meudon, France.

2 LUTH-Observatoire de Paris, CNRS, Universite Paris
Diderot, 5 place J. Janssen, 92190 Meudon, France

3 Ecole Polytechnique 91128 Palaiseau, France.

temp erature anisotropy instabilities occur (|Garv et al.l
I2001D and can remove or inject energy in the turbulent
cascade. However, for most of the solar w i nd observa-
tions, the plasm a is stable ([Matteini et al.l 120071 120111:
Bal e et alJI2009h . The energy re-distribution among the
fluid and kinetic degrees of freedom in the vicinity of
ion scales is still a matter of debate and is probably
at the origin of the spectral variations observed be-
tween 0.3 and 3 Hz in the satellite fra me: the spec-
tral index here varies between —4 and —2 (jLeamon et al.l
ll99llSmith et al.|[2006l:ISahraoui et al.ll2010D . Thisspec-
tral r ange is usually attributed to the ion dissipation
range (jLeamon et al.lll99all99ll2000t[Sm"ith et al.ll2012l)
or to another fluid cascade, which may continue down

to electron scales (Biskam p et al.l Il996t IStawicki et all
20011: ILi etall 120011: IGaltier and Bhattacharied 120031:

Galtier and Buchlid 12007^ 7"

Between ion and electron scales, the fluctuations of
the electron fluid form a sm all scale inertial range
(I Alexandrova et al.ll2007ll2008D . or, following the nomen-
clature o f ISmith et all (j2012f ). an electron inertial range.
Here, ind eed, a reproducib l e spe c trum oc kJ 2S is
observed (lAlexandrova et all 120091 : iChen et alfl2010l :
ISahraoui et al.l 120101 ). Approaching electron scales,
one may expect to observe an electron dissi pation
range, as was suggested by lAlexandroval ([2008). At
such small scales, there are only a few obs ervations
(jAlexandrova et al.l 120091 : ISahraoui et al.l 120101 ) and the
descriptions are different. Larger statistical studies are
needed to establish more firmly the properties of turbu-
lent spectra at electron scales.

In this paper we present a large statistical study of
magnetic spectra starting at ion scales and going be-
yond electron spati al scales. We use data from the
STAFF instrument (ICornilleau-Wehrlin efaTI 119971) on
the Cluster mission ([Escoubet et al.lll997l ). which is able
to measure such a range of scales. In a previous study,
lAlexandrova et all (|2009D described the electron inertial
and the electron dissipation ranges separately: a power-
law ~ fc -2 ' 8 for the inertial range and a curved spectrum
oc exp(— \fk]k~o) for the dissipation range. This model is

2

Alexandrova et al.

10"

V(km/s) e [300,700]

10

\ B(nT) 6 [2,20]

^ \ T,/T e e [0.5,5]

10

— X

^ Ns. |8e e [01,20] -

10" e

10" e

staff noise ^ ^ "

10" 10

1

10 100

frequency, f (Hz)

Figure 1. Frequency spectra with a signal to noise ratio greater
than 3 measured by Cluster— 1/STAFF in the free solar wind (for
27 intervals randomly chosen among 100). The dashed line shows
the instrument noise level. The vertical dotted line corresponds
to / = 3 Hz. The legend indicates the variations of some solar
wind parameters for the studied data set of 100 spectra: speed
V, magnetic field B, temperature ratio and the ion and electron
plasma j3.

rather complicated and has a large number of free param-
eters. In the present study, we propose a single algebraic
description for both ranges, namely, an exponential with
a power-law pre-factor: E(k±) = AkJ" exp(— k±£d)- We
find that this model describes well the totality of the ob-
served spectra at scales smaller than Ai and pi and that
its cut-off scale correlates with p e . The power- law
exponent a is found to be close to —8/3. This model
(henceforth called "the ezp-model" ) has only one free
parameter, the ampli tude of the spectrum .

Previous authors (jSahraoui et al.l I2010T ) have used a
double power-law model with a break to fit the observa-
tions in the electron inertial and dissipation ranges. We
have applied this model as well to our data, and we find
that the first pow er-law exponent is consist e nt with the
previ ous studies (|Alexandrova et all 120091 : iChen et ah!
120101) while the second exponent varies a lot. Despite
the fact that the double power-law model has more free
parameters than the exponential model used here, we
find that it describes only 30% of the observed spectra
and that the associated break scale does not present any
clear correlation with an electron characteristic scale.

2. OBSERVATIONS

For our statistical study we select homogeneous inter-
vals of 10 minutes (long enough to study kinetic scales)
within the five years interval (2001-2005) of Cluster. We
eliminate time intervals during which Cluster is magnet-
ically connected to the bow-shock by using electrostatic
wave spectrograms, which show clearly waves typical
of the electron fores hock (jEtcheto and Faucheuxl 119841 ;
lLacombe et al.lll985l). and by u s ing th e shock model de-
scribed by lFilbert and Kelloggl (|1979t ). For small angles
Qbv between the interplanetary magnetic field B and
the solar wind velocity V, Cluster is connected to the
shock. Thus, our data set only contains intervals for
which the angle Qbv > 60°. If the turbulent fluctu-
ations have a phase speed V<j> -C V, Cluster detects by
Doppler shift the fluctuations with k|| V. As B and V are

0.01 0.10 1.00

k. L (krrT 1 )

Figure 2. Fit of the most intense spectrum of Figure 1 with
the exp-model. The spectrum was measured by Cluster- 1/STAFF
on 22/01/2004. Green crosses represent the SC measurements,
red stars show the raw SA measurements without correction of
the 1st three underestimated points, visible here around 0.1km -1 .
Diamonds indicate the STAFF-SA noise level. The blue arrows
indicate inverse ion and electron Larmor radii, the black ones cor-
respond to the inertial lengths. The solid line gives the exp- model

Afc- 8 / 3 exp(-fcp e ).

quasi-perpendicular, Cluster measures fluctuations with
klB. We apply the Taylor hypothesis to get the wave-
number from the frequency, k± — 2irf /V. However,
about ~ 10% of the pre-selected intervals show the pres-
ence of right hand polarised whistlers in quasi-parallel
propagation. For these waves the Taylor hypothesis is
not applicable, because > V. We discard these in-
tervals in the present study. This data selection process
gives us 100 intervals. Within this statistical sample, the
plasma conditions vary as usually in the solar wind in fast
and slow streams at 1 AU (see the legend of Figure [lj .

Figure [1] shows the total Power Spectral Density (PSD)
of magnetic fluctuations, for 27 intervals randomly cho-
sen among 100, as a function of frequency in the space-
craft frame P(f). These spectra are measured by the
STAFF Search Coil sensor and analyzed onboard by
the magnetic waveform unit (hereafter called SC) at
/ £ [0.5, 9] Hz and by the Spectrum Analyser (hereafter
called SA) at / > 8 Hz.

The spectra are analyzed only for the frequencies where
the Signal to Noise Ratio (SNR) is larger than 3. The
spectral parts below this threshold are not shown to avoid
any erroneous interpretation. As one can see from Fig-
ure [TJ this instrumental noise limit allows us to use data
up to 30 — 400 Hz, depending on the turbulence intensity
(i.e., for the most intense spectrum, we have valid obser-
vations up to 400 Hz). The analyzed range of frequencies
corresponds to / E]f c i, fee]-

A poor calibration of the first 3 frequencies of SA (at 8,
11 and 14 Hz) [Y. de Conchy and N. Cornilleau, private
communication, 2011], was corrected by an interpolation
of these points between the highest SC frequency and
the 4th point of the SA spectra. The linear interpolation
between log 10 P(f) and log 10 / is possible as far as the
spectra follow a power-law at these frequencies. An ex-
ample of a raw spectrum without the correction can be
found in Figure O

Turbulent spectrum at kinetic scales

3

3. ALGEBRAIC DESCRIPTION OF TURBULENT SPECTRA
AT SCALES SMALLER THAN Pl AND A 7 .

3.1. Exponential model

Here we propose a model to describe the whole turbu-
lent spectrum at scales smaller than pi and Xi and down
to a fraction of the electron scales with the smaller pos-
sible number of parameters, namely an exponential with
a characteristic scale l d and with a power-law pre-factor

E(k±) = Ak~ a exp(-k±e d ).

(1)

This exp-mode\ has three free parameters: the amplitude
A, the spectral index a and the cut-off or "dissipation"
scale l d .

We start by fitting the model (p} to the 100 observed
spectra (with a signal to noise ratio > 3, as explained
in section 2) for k± corresponding to / > 3 Hz (see ver-
tical dotted line in Figure [I| , assuming that the three
parameters have independent variations.

Figure [2] gives the fit with the most intense spectrum
of Figure [T] as a function of the wave-number P(kj_) =
P(f)V/2w, which is determined using the Taylor hy-
pothesis and the energy conservation law J P(k±)dk± =
J P(f) df. Green crosses show the Morlet wavelet spec-
trum (|Torrence and Compol [l99l of STAFF-SC mea-
surements. Red stars display the STAFF-SA data for
the same time period. (In this plot we keep the 3 first
poorly calibrated data points, one can see them around
k = 0.1 km -1 and compare with the result of the in-
terpolation in Figure 1). The error bars are estimated
from the v ariance over 10 minutes of the PSD at each
frequency (|Alexandrova et al"1l2010D . This spectrum is
valid up to ~ 400Hz, which gives us the maximum wave-
vector k ~ 4 km" 1 (while l/p e — 1 km" 1 ). This is the
smallest scale ever measured with a good sensitivity at
1 AU in the solar wind. The exp-model ([T]) fitting is
shown by the black solid line. The parameters of the fit
in this case are a = 2.70±0.15 and i d = (0.90±0.25) km,
while p e = (0.95 ± 0.05) km.

Figure [3] summarizes the results of the fitting for the
100 spectra. Panel (a) shows the histogram of the spec-
tral index, a — 2.63 ± 0.15, the error being the standard
deviation of the mean. Note that (a) ~ 8/3. It appears
that the variations of a and l d are not independent since
the dispersion in a is due to the variations of the cut-
off scale l d as observed in Figure [3]b. A linear fit gives
£ d (km) = 12.9 -4.4a, i.e.

a = 2.9- V 4 - 4

(2)

i.e. if l d was small, a would be appro ximatively equal to
2.9, a value close to the one found by lAlexandrova et all
(|200l .

On the other hand, the variations of £ d are related to
the variations of the electron Larmor radius, £ d ~ 1.35p e ,
as shown in Figure [3J;, with a relatively high correlation
coefficient of 0.70. Figure [3ji shows a positive but much
weaker correlation of 0.34 between the dissipation scale
£ d and the electron inertia length X e .

The results presented in Figure [3] suggest that within
the framework of the exponential model there is only one
free parameter, the amplitude of the turbulent spectra,
A, and the observed spectra can be described approxi-

Model: E(k) = Ak~"exp(-k/k d ), k d =l d 1

6 [
5 r
4 \

6

& 3 \

Cor=-0.83
L=12. 86-4. 35a

(b)

2.0 2.5 3.0 3.5 4.0

1 ;

[

2.0 2.5 3.0 3.5 4.0

6
5
4

6

A 3
- 2
1

Cor = 0.70

i d = -0.36+1.35/),

•«««•

5 ■
~ 4 ■

6

A 3

ts

- 2
1

I

. . • • • •

0.0 0.5 1.0 1.5 2.0 2.5 3.0
p e (km)

0.0 0.5 1.0 1.5 2.0 2.5 3.0
K (km)

Figure 3. Results of the fitting with the exp-model for the 100
observed spectra: (a) histograms on the spectral index a; (b) cut-
off scale id as a function of a; (c) as a function of the electron
Larmor radius p e and (d) 1$ as a function of the electron inertial
length A e .

mately by

E(k±) ~ Ak ± 8 ^ 3 exp(— k±p e )

(3)

We verify this point in Figure 01 where we superpose
the 100 spectra analyzed here with the 7 spectra cov-
ering fluid and kinetic scales from ([Alexandrova et aLl
2009). The spectra are shifted vertically by a param-
eter Eq (a re lative spectral level) , in the same way as
in Fig. 2 of ([Alexand rova et all I2009D . The superpo-
sition of the 107 spectra is nice, which indicates the
generality of the turbulent spectrum E(kj_) in the so-

— 5 /3

lar wind: it follows oc k ± at MHD scales and cx

fcj^ 3 exp(— k±p e ) at scales smaller than the ion kinetic
scales Xi and pi (i.e. kp e > 0.03). The bottom panel
shows the 100 spectra E(k±p e ) compensated by a func-
tion F = (kj_p e ) 8 / 3 exp(k±p e ) for kp e > 0.03: the re-
sulting spectra are flat, indicating that the exp-model
with one free parameter describes well all the turbulent
spectra in the solar wind at these scales and is valid for
nearly 2 decades in wave numbers. Note that a damp-
ing length i d variation of more than 20% with respect to
the mean p e values results in a strong departure of the
compensated spectra.

The amplitude A of the turbulent spectra (related to
the parameter Eq) is found to be correlated with the
ion thermal pressures nfcT,, as w ithin the MHD range of
turbulence (jGrappin et al.l [19901 ) . and with the ion tem-
perature anisotropy Ti±/Tq (not shown). Other plasma
parameters seem to be less important, but still it is im-
possible to exclude completely the influence of the mag-
netic and kinetic energies in the solar wind (paper in
preparation).

3.2. Break model

Is there another simple model which represents well
the observations with a small number of free parameters?

4

Alexandrova et al.

107 solar wind spectra

10~* 10" 1 10° 10 1

k x (km -1 )

compensated spectra

10

x 8/3 exp(x)E(x)
E j jfrB W it t | S H ~*

0.1

1.0

x = kjp e

Figure 4. Upper panel: 107 superposed solar wind spectra, the
100 spectra in the kinetic range analysed in th is paper and the 7
spect ra covering fluid and kinetic scales from fAlexandro va et alj
120091 1. the constant being Eq G [0.4,95]. All spectra have the
signal to noise ratio greater than 3, as in Figure [T] The blue line

indicates oc fc^ 5//3 and the red line shows oc fc^ 8//3 exp(— kj_p e ),
that represents well all the spectra. Bottom panel: the diamonds
show 100 compensated spectra E(k±p e )- F(k±p e ), with F(k±p e ) =
(fc±pe) 8//3 exp(fcxPe), for k±p e > 0.03. The solid line denotes the
most intense compensated spectrum.

Let us compare the turbulent spectra within the electron
inertial and dissipation ranges (i.e. scales smaller than
Pi and Xi) with the double power-law or break-model

E(k±) = Axk± ai (1 - H(kx - k b )) + A 2 k~ a2 H(k±

h),

(4)

H{k± — kb) being the Heaviside function, kb the wave
number of the break scale lb = l/kb, A\_ 2 the amplitudes
of the two power-law functions with spectral indices a\^ 2
on both sides of kb- This model has five free parameters.
Note that equation ((4]) is not diffcrentiable for k = kb-
Near kb the turbulent level has to be the same on both
sides. Thus, kb can be determined by the four other
parameters of the model

logio k b =

\og w (A 1 /A 2 )
ai - a 2

(5)

Iterations to find the model parameters with condition
(O converge to different results, depending on the initial
kb- Therefore, minimising the error of the fit over kb
is needed, so that finally the model has still five free
parameters.

We apply this fereafc-model to the 100 observed spectra
within the same /c^-range as was done for the exp-model.
Despite the fact that the &reofc-model has more free pa-

rameters than the exponential model, we find that it can
be applied only to 30 spectra; for the 70 other spectra
there is no solution with condition ([5]) verified, indicat-
ing the absence of a clear break point, or, for a part of
these spectra, not enough data points to isolate the sec-
ond power-law.

Model: E(k) = A 1 k~ a '(l-H(k-kJ)+A,k~ a 'ri(k-k b ), k b =l b ~'

2.5 3.0 3.5 4.0 4.5 5.0

a,, a.

4

3

?

1

(c)

4

3

?

1

4 f
3 ■

?

1

Cor, = -0.57

Cor 2 =-0.58 ,

•

(b)

2.5 3.0 3.5 4.0 4.5 5.0

a,, a ?

(d)

0.0 0.5 1.0 1.5 2.0
p e (km)

I 1

0.0 0.5 1.0 1.5 2.0 2.5
K (km)

Figure 5. same format as Figure[3] but for the parameters of the
break-model determined from the fitting with 30 spectra, i.e. (a)
histograms on the spectral indices; (b) break scale £{, as a function
of Qi (open circles) and 02 (filled circles); (c) it, as a function of the
electron Larmor radius p e and (d) if, as a function of the electron
inertial length A e .

Figure [5] summarizes the parameters of the break-
model determined from the fitting of the 30 spectra. For
these spectra, the condition © is verified. Panel (a)
shows histograms of the spectral indices: the mean val-
ues are ax = 2.86 ± 0.08, a 2 = 3.91 ± 0.29. Note the
narrow dispersion of the spectral index a±. It is close
to a, when £d is negligible (see eq. ([2])) so confirming
the spectrum between io n and electron scales found by
lAlexandrova et all (|2009f ). So we can fix one of the pa-
rameters of the model. The second exponent a 2 has
a large dispersion, not found to be controlled by any
plasma parameter. The values of the spectral indices are
correlated to the position of the break scale lb (see panel
b). Figure [SJ: shows lb as a function of p e . No correla-
tion is observed. Figure [SJl shows lb as a function of A e :
the correlation is positive but weak (~ 0.43). From the
comparison of the observed turbulence spectra at plasma
kinetic scales with the &reafc-model, one may conclude
that this model has one fixed parameter, another is fixed
by the condition ([5]) , the other 3 parameters are free, not
found to be determined by plasma parameters.

It is important to note that the errors of the fit of the
break and exp-models are of the same order. Figure [6]
shows an example of a solar wind spectrum (black lines)
fitted with both models, with the least mean square dis-
tance between the observed spectra and the fit, A 2 , given
in the caption; the insert shows histograms of £\ 2 exp (red)
for 100 spectra and A 2 reafe (blue) for 30 spectra: the
same mean values of the histograms are observed. This

Turbulent spectrum at kinetic scales

5

0.01 0.10 1.00 10.00

k x (km" 1 )

Figure 6. Example of an observed spectrum (lower black line)
compared with E(k) ~ fc~ 2 - 7 exp(— kl d ) (red dashed line), with
£^ marked by a red diamond; the error of the fit is A 2 ,^ =
1.8 X 10" 3 (nT 4 km 2 ). The same spectrum, shifted by a factor
10 2 and compared with E(k) ~ fc~ 2,8 below the break shown by a
blue diamond, and with ~ k~ 40 above it (blue dashed line); here
A 2 reafc = 1.9 X 10 — 3 (nT 4 km 2 ). Ion and electron characteristic
scales are marked by arrows; a vertical dotted line indicates the
beginning of the fc-domain where the fits are performed. Insert:
histograms of the fitting errors, A§™ (red, for 100 spectra) and

^break ( bme i for 30 spectra).

leaves us free to choose the model, based on other crite-
ria than goodness of the fit, namely, on the number of
degrees of freedom of the model and on the number of
described cases.

4. DISCUSSION AND CONCLUSION

The exp-model E(k±) — Ak ± S ^ 3 exp(— k±p e ) proposed
in this study provides a single algebraic description of
the solar wind spectrum at scales smaller than the ion
characteristic scales, Xi and pi, and going beyond the
electron scales (i.e. within the electron inertial and dis-
sipation ranges). This model describes well the totality
of the observed spectra and has only one free parame-
ter - the amplitude A of the spectrum. The amplitude
seems to be a function of the ion thermal pressure and
the ion temperature anisotropy in the solar wind. How-
ever, it is difficult to exclude the role of the magnetic and
kinetic energies: more work is needed to determine the
exact relationship between the amplitude of the turbu-
lent spectrum and the energy budget in the solar wind.

The spectral index close to —8/3 observed in the solar
wind at scales smaller than ion characteristic scales is
in agreement with quasi-bidimentional strong Electron
MHD turbulence (k± ^> kn) when parallel cascade is
weak (jGaltier et al.l 120051) . Recently, the same spectral
index was found as well in stron g kinetic Alfven turbu-
lence (Boldvrc v and Per ez 2012).

In usual fluid turbulence, the far dissipation range
is described by E (k) ~ fc 3 exp(— ckld) (with c ~ 7)
(jChen et al.l 119931) . This is due to the resistive damp-
ing rate oc k 2 valid in a collisional fluid, which gives an
exponential spectral tail. In the collisionless plasma of
the solar wind there is no resistive damping, and thus

t his coincidence dese rves an explanation.

Howes et al.1 (|2011f) consider a model ("weakened cas-
cade model") which includes the nonlinear transfer of
energy from large to small scales in Fourier space and
the damping of kinetic Alfven waves (KAW's). For com-
pleteness, we discuss now this model in some detail. The
model reads for the magnetic energy b 2 at scale k:

d t b\ = -k ± d k e - 2 7 b 2 + S (6)

with e being the magnetic energy transfer rate, S being
the source term. The damping term 2~fb 2 is obtained by
linearizing the Vlasov-Maxwell equations in the gyroki-
netic limit (fcy -C kx, with frequencies / <C /«)■ In the
limit kpi ^> 1, this gives

7 ~fc||K(fc±p l ) 2 , (7)

see eq. (63) in (jHowes et al.l I2006D . To complete the
model, we must write down the expression for the mag-
netic energy flux e. This is given by

e = b 2 /r = k ± ub 2 (8)

where r = l/k±u is the nonlinear time and u the veloc-
ity fluctuation. At MHD scales k±pi < 1, we have for
Alfven waves u ~ b, but at small scales k ±pj > 1, we
have for KAW u/b ~ kj_p i7 see eq. (3) in (How es"et al.l
1201 If ). Using these expressions in eq. © and (jH}, one ob-
tains scaling laws for the magnetic energy spectrum as
stationary solutions of the transfer equation ©, neglect-
ing the damping and source terms. The spectral laws are

respectively Ek oc kj 5 ^ 3 at large scales and Ek oc k ± 7 ^ 3
between ion and electro n scales. When tak ing into ac-
count the damping term, IHowes et al.l (|2011D obtain nu-
merically the same spectral laws, with a final curved tail
at scales smaller than electron scales. Superficially, this
spectrum thus resembles the analytic form which we have
found to be valid to describe the solar wind turbulence.

We last rema r k tha t the damping term in the model
of IHowes et all (|2011D for kxpi > 1 is (see eq. (7))
of the form 7 oc fciifc^. Taking into account the as-
sumption of critical balance r = ta (i.e. k±u = fen Va)
(|Goldreich and Srid har 1995), with ta the Alfven time
and Va the Alfven speed, and the spectral index —7/3

(i.e. u ~ k^ 2 ^ 3 ), one gets k» oc k 1 ^ 3 . Therefore, the

damping term takes the form 7 oc . The exponent

of the damping rate is thus very close to the k 2 scaling
of the Laplacian viscous term, which is known to lead in
hydrodynamical turbulence to an exponential tail in the
dissipation range.

This model does not take into account the cyclotron
damping. So, while the proposed phenomenology may
explain the exponential tail of the fc^-spectrum stud-
ied here, it cannot describe more isotropic wave vectors,
which might be present as well in the solar wind. It is
possible that the dissipation mechanism could also be
due to electron-cyclotron absorption of oblique short-
wavelength whistler waves, or even of lower- hybrid waves.
More observations under different field-to-flow angles
Qbv ar e needed within the electron inertial and dissi-
pation ranges to address this point.

To build a realistic model of the dissipation in the
solar wind we need to resolve still an open question

6

Alexandrova et al.

till

O

w

.00 0.01 0.02 0.03 0.04 0.05
d(V)/V

0.0

0.1
d(

0.2

0.3

Figure 7. Histogram of the variability of the solar wind speed V during 10 minutes, for a sample of 100 intervals (left panel). Right panel
: histogram of the variability of the electron gyroradius p e .

on the nature of the turbulent fluctuations. Some au-
thors argue that th e electron inertial r ange is a whistler
mode turbulence (jSaito et al.l 120101 : iNarita and Gary
20101) . others suggest KAW t urbulence (IBale et al "

2005t ISchekochihin et all 120091 ISahraoui et al.l 120101:
Salem et al.l 120121) . or a combination of both types of

linear waves dPodesta and Gary! l201ll) . The model of
I Howes et al.l ()2011l ). described above is based on KAW
turbulence as well. However, it is still not clear whether
we can describe turbulence in the solar wind as a mixture

of linear waves (weak turbulence) which will dissipate ho-
mogeneously in space (or in the plane perpendicular to
B), or if it is a strong turbulence with dissipation re-
stricted to intermittent coherent structures. What is the
topology of these structures - current sheets, shocks or
coherent vortices?

In the present study we have limited ourselves to ob-
servations of the spectral shape in the electron inertial
and dissipation ranges. Our results give observational
constraints for future theoretical models.

APPENDIX

PLASMA PARAMETERS VARIATIONS DURING THE SPECTRA INTEGRATION TIME

One could argue that the exponential bending found in the fc-spectra of the magnetic fluctuations is due to variations
of the solar wind speed V or of the gyroradius p e during the 10 minutes of each considered interval. Indeed, each
fc-spectrum is obtained with an average P(f) over 150 frequency spectra, itself shifted in the fc-domain with the average
V. The standard deviation dV over 10 minutes is very small. Figure Eta) displays the histogram of the ratio dV/V:
91% of the 100 considered intervals have dV/V < 0.02. Thus, the shift of P(f) in the k-domain with the average V
cannot change the spectral bending. Similarly, the standard deviation d(p e ) over 10 minutes is small. Figure EJb)
displays the histogram of the ratio d(p e )/p e : 96% of the 100 considered intervals have d(p e )/p e < 0.1. Thus, the use
of the average p e for an interval, in place of the exact p e for each of the 150 k-spectra, cannot produce the observed
bending which covers a wide range of fc, and cannot smooth a possible spectral break.

LOG-SPACED FREQUENCIES OF THE CLUSTER/STAFF-SA INSTRUMENT

Another argument against the observed spectral bending could be that it is an artefact due to the logarithmic
frequency binning of the Spectrum Analyser (SA) on Cluster.

Indeed, the centre frequency of the output channels of SA are distributed logarithmically (between 8.8 Hz and 3.56
kHz), and each channel has a bandpass proportional to its centre frequency, 2df = 26//100. As the onboard waveforms
used by SA are lost, we have no way to check whether a different frequency binning would give different frequency
spectra. We only make a crude comparison between an analytical frequency spectrum Pa(I) and the same spectrum
integrated by a trapeze method in the logarithmic frequency bands of SA, Prif)- Note that the trapeze integration has
to be made on the logarithms of Pa (/) because the gain of the STAFF-SA receivers is proportional to the logarithm
of the power. Figure [8] gives the results of this comparison for a spectral break model (with two spectral indices
a.\ = 2.8, a 2 = 4, left panel) and for an exponential model (right panel) Pa(J) = f~ 2 ' 7 cx p(— ///())> with f ~ 115 Hz.
The solid lines give -Pa(/), and the crosses give Pr(f) m the 18 lowest frequencies of SA. It is clear that the trapeze
integration in logarithmic channels does not change the shape of the spectra. The ratio R = Pt/Pa is very close
to unity: 1.02 to 1.03. There is thus a slight systematic overestimation of the spectrum by the trapeze integration
in logarithmic bands. This overestimation (2 to 3%) cannot increase the downward bending of the spectrum. We
conclude that the logarithmic frequency binning of the Spectrum Analyser cannot smooth a possible spectral break,

Turbulent spectrum at kinetic scales

7

break model exponential model

10 100 10 100

f (Hz) f (Hz)

Figure 8. Comparison between an analytic power spectrum Pa(J) (solid line) and the power spectrum Prif) (crosses) calculated by a
trapeze integration of Pa(I) with the logarithmic frequency binning of the Spectrum Analyser (2df = 26%), for the two models discussed
in the paper. The difference between Pt{I) and Pa(I) is less than 3%.

and cannot produce an artificial downward bending of the spectra.

REFERENCES

O. Alexandrova. Solar wind vs magnetosheath turbulence and Alfven vortices. Nonlinear Processes in Geophysics, 15:95-108, Feb. 2008.
O. Alexandrova, V. Carbone, P. Veltri, and L. Bprrisn-Valvn Snlar wind Cluster observations: Turbulent spectrum and role of Hall effect.

Planet. Space Sci., 55:2224-2227, Dec. 2007. |doi:10 1016/j.pss. 2007.05. 022| .
O. Alexandrova, V. Carbone. P. Veltri, and I,. So rriso- Valvo. Small-Scale Energy Cascade of the Solar Wind Turbulence. Astrophys. J.,

674:1153-1157, Feb. 2008. |doi:10. 1086/524056}
O. Alexandrova, J. Saur, C. Lacombc, A. Mangeney, J. Mitchell, S. J. Schwartz, and P. Robert. U niversality nf Solar- Wind Tnrhnlcnt

Spectrum from MHD to Electron Scales. Physical Review Letters, 103(16):165003— h Oct. 2009. |doi: 10. 1 103 /PhysRcvLctt. 103. 165003 .
O. Alexandrova, J. Saur, C. Lacombe, A. Mangeney, S. J. Schwartz, J. Mitchell, R. Grappin, and P. Kobert. Solar wind turbulent

ap^r-trnm frr,m MUT~)

o electron scales. 12th Int. Solar Wind Conference, AIP Conference Proceedings, 1216:144-147, Mar. 2010.

doi:10.1063/l. 3395821

. D. Bale, P. J. Kellogg, F. S. Mozer, T. S. Horbury, and H. Reme. Measurement of the Elec tric Fluctuation Spectrum of

June 2005. Hoi:10.1103/PhysRevLett.94.215002

Magnetic Fluctuation Fower Near Froton Temperature

Magnctohydrodynamic Turbulence. Physical Review Letters, 94(21):215002
. D. Bale, J. C. Kasper, G. G. Howes, E. Quataert, C. Salem, and D. Sundkvist
Anisnt.rnpy Instability Thresholds in the Solar Wind. Physical Review Letters, 103(21):211101— h Nov. 2009
doi:10.1103/PhysRevLett. 103.211101
). Biskamp.

E. Schwarz,
1264-1267, Feb. 1996.
Boldyrev and J. C.
Bourouaine, O.

md I E I h-ql-f

Physical Review Letters, 76:

Twn-nimenQinTial Electron Magnetohydrodynamic Turbulence.

doi: 10. 1 103/PhysRcvLctt .76. 1264 .

Pcicz. Spectrum of Kinctic-Altvcn turbulence. ApJ, 758:L44, Oct. 2012. |doi: 10. 1088/2041-8205 /758/2 /L44 .
Alexandrova, E. Marsch, and M. Maksimovic. On ^n^tral RroAg in ilio Pr,woV- sip^ira CTagnetlc Fluctuations in
Fast Solar Wind between 0.3 and 0.9 AU. ApJ, 749:102, Mar. 2012. |doi:10.1088/0004-637X/749/2/102 .
C. H. K. Chen, T. S. Horbury, A. A. Schekochihin, R. T. Wicks, O. Alexandrova anrl .1 Mit chell Anisntrnpy of Solar Wind Turb ulence

between Ion and Electron Scales. Physical Review Letters, 104(25):255002— (-, June 2010. doi:10.1103/PhysRcvLctt. 104.25500

Chen, G. Doolen, J. R. Herring, R

Letters, 70:3051-3054, May 1993. |doi: 10 1 103 /PhysRcvLctt.70.3051

. Cornilleau-Wehrlin et al. The Cluster Spatio- Temporal Analysis of Field Fluctuations (STAFF) Experiment
79:107-136, Jan. 1997.

H KVairhnan S A ("irQ^ag anrl Z. S. She. Far-dissipation range ot turbulence. Physical

Space Science Reviews,

CP F, S mnhnt. Ft Schmidt, an

doi:10.1023/A:1004923124586
. Etchcto and M. FaucKeuie I

M. L. Goldstein. Cluster - Science and Mission Overview. Space Science Reviews, 79:11—32, Jan. 1997.
tailed study nf electron plasma waves upstream of the earth's bow shock. J. Geophys. Res., 89:

lauchc

6631-6653, Aug. 1984. Hoi: 10. 1029/ JA089iA08p06631
C. Filbert and P. J. K ellogg H'lcrtrostatie noise at. T ie plasma frequency beyond the earth's bow shock. J. Geophys. Res., 84
1369-1381, Apr. 1979. |doiT0.1029/JA084iA04p01369[

The legacy of A.N. Kolmogorov. Cambridge: Cambridge University Press, 1995.

Frisch. Turbulence.
Galtier and A. Bhattacharjee
3065, Aug. 2003.

Haltier and F, Bnc hlin. Multiscale Hall-Magnetohydrodynamic Turbulence in the Solar Wind

Anisotropic weak whistler wave turbulence in electron magnetohydrodynamics. Physics of Plasmas, 10:

Astrophys. J., 656:560-566, Feb. 2007.

doi:10.1086/510423

S. Ualtier, A. Fouquet, and A. Mangeney. On spectral sealing laws fnr inrnmpressihle anisotropic magnetohydrodynamic turbulence.

Physics of Plasmas, 12(9):092310, Sept. 2005. |doi:10.1063/l. 205250?] .
S. P. Gary, R. M. Skoug, J. T. Steinberg, and C. W SmTEE FrpES temperature anisotrgpy constraint in the solar wind: ACE

observations. Geophys. Res. Lett., 28:2759-2762, July 2001. doi:10.1029/2001GL013165

Alexandrova et al.

P. Goldreich and S. Sridhai

763-775, Jan. 1995. |doi:10.1086/175121
R. Grappin, A. MangcnoyZanS

8197-8209, June 1990. loi:10.1029/JA095iA06p08197
G. G. Howes, S. C. Cowley~"W

TmnarA a thor.ry of interstellar turbulence. 2: Strong alfvenic turbulence
On the origi n of solar wind MHD turbulence

and A A SrliplfocliiViin Astrophysical Gyrokinetics: Basic

The Astrophysical Journal, 438:
HELIOS data revisited. J. Geophys. Res., 95:

G

R

R

Physics of Plasmas,

Uorland, (_;. W. riammett, E. Qualaexl.
Equations and Linear Theory. ApJ, 651:590-614, Nov. 2006. |doi: 10. 1086/506172
G. Howes, J. M. Tenharg e and W. Dorla.nd A weakened cascade model tor turbulence in astrophysical plasmas.
18(10):102305, Oct. 2011. |doi:10 1063/1.3646400[
N. Kolmogorov. The local structure ot turbulence in incompressible viscous fluid for very large Reynolds numbers. Dokl. Akad. Nauk
SSSR, 30:9-13, 1941.

Lacombe, A. Mangeney. C C Hamey, and T Scudder Electron Plasma Waves Upstream of the Earth's Bow Shock. J. Geophys.

Res., 90:73-94, 1985. |doi:10.1029/JA090iA01p00073 .

J. Leamon, C. W. Smith, N. F. Ness, W. H. JVlatthaeus, and H. K. Wong. Observational constraints on the dynamics of the
interplanetary magnetic field dissipation range. J. Geophys. Res., 103(12):4775, mar 1998.

J. Leamon, C. W. Smith, N. F. Ness, and H. K. Wong. Dissipation rangn dynamic- T<Tinr-ti.~ Alfiron waves and the importance of

1999
Mullah"

electron beta e - J- Geophys. Res., 104(13):22331-22344, Oct
R. J. Leamon, W. H. Matthaeus, C. W. Smith, G. P. Zank, D. J
Wind and Corona. The Astrophysical Journal, 537:1054

H. Li, S. P. Gary, and O. JSiaaricki On thn diQQipatipn of magnetic fluctuations m the solar wm>

1347-1350, Apr. 2001. hoi:10.1029/2000GL012501 .
L. Matteini, S. Landi, P. Hellinger, F. Pantellini, M. Maksimovic, M. Velli, B. E. Goldstein, and E Marsch Fvolutinn of the solar wind

proton temperature anisotropy from 0.3 to 2.5 AU. Geophys. Res. Lett., 34:L20105, Oct. 2007. doi:10.1029/2007GL030920.
L. Matteini, P. Hellinger, S. Landi, P. M. Travmcek, and M Velli Ton Kinetir-c in the Solar Wind: Coupling Global Fxpansion to Local

doi:10 1029/1999JA900158
an H S Oiight.nn MHII -Hriven Kinetic Dissipation in the Solar
1062, July 2000. |doi:10i086/309059|.

1 Geophysical Research Letters, 28:

Y

Microphysics. Space Sci. Rev., page 128, Apr. 2011.
Narita and S P. Oary. Tncrt.ia.l-ra.nge spectrum of w

doi:10.1007/sll214-011-9774-
listler turbulence. Annates Geophysicae, 28:597-601, Feb. 2010.

doi:10.5194/angco-28-597-2010.

J. J. l-'odcsta and B. f. Gary. Magnetic Helicity Spectrum of Solar Wind Flni-tnatinnc ac a Function of the Angle with Respect to the

do i:10.1088 /0004-637X/734/l/l, 1
Kezcau.

Local Mean Magnetic Field. ApJ, 734:15, June 2011
F. Sahraoui, M. L. Goldstein, G. Belmont, P. Canu, and L. Kezeau. three Dimensional Anisotn

Subproton Scales in the Solar Wind. Physical Review Letters, 105(13):131101 — h, Sept. 2010
S. Saito, S. P. Gary, and V Narita Wavennmber spectrum of whistler turbulence: Particle-in-ce

C

(12):122316, Dec. 2010.
S. Salem, G. G. Howes,

doi:10. 1063/1. 3526602
T7

opi

efn

oi:10.1103/PhysRevLett. 105. 131101
simulation. Physics of Plasmas, 17

aer. Identification of Kinetic Alfvcn

Sundkvist, S. D. Bale, C. C. Qha^n n H if nhon and T7 s Maz
Wave Turbulence in the Solar Wind. ApJ, 745:L9, Jan. 2012. |doi: 10. 1088/2041-8205 /745/1 /L9 .
A. A. Schekochihin, S. C. Cowley, W. Dorland, G. W. Hammett, G. G. Howes, E. CJuataert, and T. Tatsuno. Astrophysical Gyrokinetics:
.Kinetic and Fluid Tnrhnlent Cascades in Magnetized Weakly Collisional Plasmas. Astrophys. J. Suppl. Ser., 182:310-377, May 2009.
doi:10.1088/0067-0049/182/l/3ic| .

C. W. Smith, K. Hamilton, B. J. Vasquez, and R. J. Leamon. Dependence of the Dissipation R ange Spectrum of In terplanetary Magnetic
Fluctuationson the Rate of Energy Cascade. Astrophys. J. Letters, 645:L85-L88, July 2006. doi:10. 1086/506151.

C. W. Smith, B. J. Vasquez, and J. V. Hollweg QJjsenatiqiiaJ Prmctraintc nn the Role of Cyclotron Damping and

in the Solar Wind. ApJ, 745:8, Jan. 2012. |doi:10.1088/0004-637X/745/l/8
O. Stawicki, S. P. Gary, a,nri H. Li. Solar wind magn

C

Kinetic Alfven Waves

ictic fluctuation spectra: Dispersion versus damping. J. Geophys. Res., 106:

Stawicki, S. P. Gary, and

8273-8282, May 2001. ioi:10.1029/2000JA000446
. Torrence and G. P. Compo. A Fractical Guide to Wavelet Analysis. Bulletin of the American Meteorological Society, 79:61-78, Jan.
1998.

We thank the referee for constructive comments. We are also grateful to G. Howes for useful indications. We
thank the team of Cluster Active Archive. This work was partly supported by the Centre National d'Etudes Spatiales
(CNES) and Program National Soleil-Terre (PNST/INSU).